Is it possible to ‘compute conversation’? I mean — is it possible to write heuristics that respond in surprising and stimulating ways, enabling a back-and-forth exchange among intelligent participants?
Gordon Pask not only thought so, he designed and built a series of machines that did so, as far back as 1953. Sure, some of those conversations were simple and his machines could not understand language or listen to speech — but all the mechanisms he built had memory (and so could learn) and displayed novel, unexpected behaviors (and so kept their co-participants engaged in the interaction). As a result, all of Pask’s machines could hold a form of conversation with humans (or sometimes, with other machines!).
The photo above is a control panel from 1958: Pask’s ‘Eucrates‘ environment, where a ‘teacher’ machine attempts to train a ‘pupil’ machine comprising neural nets (yup, neural nets in 1958). There are knobs labelled ‘awareness’ and ‘obstinacy’ — and don’t miss the wackiest one, ‘oblivesence.’ (Pask was British, so I hold to the definition in British dictionaries: ‘willful forgetfulness.’)
Wait… what? Continue reading “Computing Conversation — A Lecture”